BOREAL ENVIRONMENT RESEARCH 21: 1–24
ISSN 1239-6095 (print) ISSN 1797-2469 (online)
© 2016
Helsinki 8 April 2016
Operational correction of daily precipitation measurements in
Finland
Antti Taskinen* and Kimmo Söderholm
Finnish Environment Institute (SYKE), P.O. Box 140, FI-00251 Helsinki, Finland (*corresponding
author’s e-mail: [email protected])
Received 19 Nov. 2014, final version received 8 Sep. 2015, accepted 8 Sep. 2015
Taskinen A. & Söderholm K. 2016: Operational correction of daily precipitation measurements in Finland. Boreal Env. Res. 21: 1–24.
An operational method for daily precipitation observation correction of aerodynamic, wetting and evaporation errors is presented. The aerodynamic correction method used has been
developed for Finnish conditions when wind observations at the gauge are not available
but the data from the nearest synoptic station can be used. Daily precipitation observations
in Finland and its cross-boundary watersheds from the period 1961–2011 were corrected.
It was discovered that the mean annual total sum of all gauges increased by 13.6% from
590.3 mm to 670.6 mm. The average overall proportions were 8.6%, 22.8% and 68.6%
for evaporation, wetting and aerodynamic corrections, respectively. On a daily basis, the
correction factor varied remarkably according to the form and intensity of precipitation,
the median value being 1.21. Positive linear trends were identified in observed, corrected
total and liquid precipitation, whereas a negative linear trend was obtained in corrected
solid precipitation. The preliminary assessment using a watershed model indicated that the
amount of correction is approximately correct for liquid precipitation but might slightly
be too large for solid precipitation. Using corrected precipitation as a hydrological model
input is recommended but a more comprehensive research is needed.
Introduction
The starting point of all hydrological considerations related to water balance studies is the
knowledge of the amount and distribution of
precipitation with respect to time and space.
This information has almost solely been based
on point measurements obtained from a cantype gauge elevated above the ground. To obtain
better information for spatial variation of precipitation, gauges have been distributed over an
area and bound into networks (Sevruk 1986b).
In recent decades, remote sensing techniques
have improved but gauge networks still play
Editor in charge of this article: Harri Koivusalo
an important role as an independent source of
precipitation information and also as a reference tool for weather radars (e.g. Collier 1986a,
1986b, Gjertsen and Dahl 2001, Saltikoff et al.
2015) and satellites (e.g. Bell and Kundu 2003).
Gauge precipitation measurements are always
subject to various sources of errors. Sevruk
(1986b) divided the sources as follows: (1)
random and systematic errors of point precipitation measurements, (2) random error at the gauge
site (due to local irregularities of topography and
micro-climatic variations), and (3) random error
of a gauge network (due to inadequate network
density). Systematic error is believed to be the
2
most important source and it can be further
divided into the following components: (1) aerodynamic, (2) wetting, (3) evaporation, (4) splash
in and out, and (5) blowing and drifting snow.
According to Førland et al. (1996), the most
important of these for the Nordic countries is the
aerodynamic error, but wetting and evaporation
errors also have to be accounted for.
Korhonen (1944) and Dahlström (1970a,
1970b) were among the first to publish precipitation error considerations and correction studies
for Nordic conditions to be followed later by e.g.
Allerup and Madsen (1980), Tammelin (1984),
Solantie (1986) and Dahlström et al. (1986). A
pioneering and influential international workshop on the topic was held in 1985, resulting in
a research report of a number of papers (Sevruk
1986a). In the same year, World Meteorological
Organization (WMO) initiated a solid precipitation measurement intercomparison to evaluate
national methods of measuring solid precipitation against methods with familiar accuracy and
reliability, including past and current procedures,
automatic systems and new methods of observation (Goodison et al. 1989). This brought about a
number of investigations during the next twenty
years or so. Some preliminary results were presented by Sevruk (1989a) and more profound,
for instance, by Huovila et al. (1988) and Førland et al. (1996), who reported on the efforts
of the Nordic countries to standardise precipitation correction methods and compare rain gauge
types. Yang et al. (1995) showed comparisons
between the Tretjakov and a double fence intercomparison reference (DFIR) in seven countries.
Allerup et al. (1997) presented a statistical model
for solid and mixed precipitation correction due
to wind-induced errors. Yang et al. (1998) investigated the effects of environmental factors on
the gauge catch and found that wind speed was
the most important one. Moreover, Førland and
Hanssen-Bauer (2000) evaluated results from the
intercomparison project and parallel precipitation measurements of Tretjakov and Norwegian
gauges and Yang and Ohata (2001) reviewed
methods to determine bias correction terms of
Tretjakov gauges and applied selected ones in
Siberian regions for 1986–1992.
In the same continuum of studies, Bogdanova
et al. (2002) presented the bias method for the
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
standard Tretjakov gauge. The method accounts
for the following corrections: (1) aerodynamic
error, (2) joint effect of wetting, evaporation and
condensation at the gauge collector interior, (3)
trace precipitation, and (4) false precipitation
(blowing snow flux into the gauge). Michelson (2004) implemented a statistical Dynamic
Correction Model with the intent to perform
a systematic correction of precipitation measurements from gauges found in and near the
Baltic Sea’s drainage basin. He found out that
H&H-90 gauge underestimates precipitation by
around 8% on average and its average correction factor is slightly smaller than that of the
Tretjakov gauge’s. Chvíla et al. (2005) showed
that the wind-induced loss of liquid precipitation increases with increasing wind speed and
decreasing intensity of precipitation. Sugiura et
al. (2006) carried out an intercomparison of
solid precipitation measurements to examine the
catch characteristics of five different precipitation gauges, including Tretjakov. They found,
for instance, that the daily catch ratios (observed
precipitation/corrected precipitation) decreased
more rapidly with increasing wind speed. Other
investigations of challenges in measuring solid
precipitation have been presented by e.g. Nitu
and Wong (2010), Rasmussen et al. (2011) and
Wolff et al. (2013). Sevruk et al. (2009) summarized the results of international precipitation measurement intercomparisons of WMO
between 1955 and 2008.
The Finnish Environment Institute runs an
automatic, operational hydrological watershed
model called the Watershed Simulation and Forecasting System (WSFS). The system is described
by Vehviläinen et al. (2005) and is available
at
http://www.environment.fi/waterforecast.
WSFS is used for e.g. flood forecasting, realtime monitoring and climate change research. It
covers the whole land-area of Finland in addition
to the cross-boundary watersheds and its time
step is one day. The water quality component
VEMALA (Huttunen et al. 2016) simulates the
transport of suspended solids, total phosphorus
and total nitrogen. Main meteorological inputs
are precipitation and temperature. This study
was triggered by the need to improve the accuracy of the precipitation input and make the
observations more spatially representative and
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
temporally consistent by accounting for the main
errors related to the gauge measurements, keeping in mind the operational nature — each observation is automatically corrected before usage
— of the method application.
The main purpose of this article is to describe
in a detailed fashion the method for correction
of precipitation measurements consisting of the
exposure method for aerodynamic correction in
addition to wetting and evaporation corrections.
The method was applied to every precipitation observation in Finland made between 1961
and 2011, using correction coefficients depending on gauge exposure information and type,
both changing in time. Moreover, the aim was
to check the data and the correction approach
results for the purpose of operational hydrology
by presenting the amount of increased precipitation at each gauge in different forms and at different time scales at each gauge keeping in mind
the spatial aspects of Finnish weather conditions.
Also the consistency of data was analysed for
climate or other research utilizing long time
series.
Climate and hydro-meteorological
observations
Climate
Finland is located between 60° and 70° of northern latitude, a quarter of the country located
north of the Arctic Circle. Finland lies between
the Scandinavian mountains and the northern
Russian plains. In the west and the south, it has
a long coastline with numerous islands along the
Baltic Sea (Finland’s Fifth National Communication under the United Nations Framework
Convention on Climate Change 2009). Due to its
location, the climate of Finland has features of
both marine and continental climates, depending
on the direction of airflow. The mean temperature in Finland is many degrees higher than in
most areas at the same latitudes. This is mostly
due to airflows from the Atlantic warmed by the
Gulf Stream but also because of the Baltic Sea
and abundant inland waters. According to the
Köppen-Geiger climate classification for Europe
(Peel et al. 2007), Finland belongs to the class
3
of cold climate without dry seasons and to the
sub-class of cold summers with the exception
of southern parts belonging to the sub-class of
warm summers.
Observing hydro-meteorological
variables
Precipitation
The meteorological data were mainly provided by
the Finnish Meteorological Institute (FMI). Since
cross-boundary watersheds were included in the
study, data provided by Swedish Meteorological and Hydrological Institute, Norwegian Meteorological Institute and Hydrometeorological
Centre of Russia were also used. In 1961–2011,
there were three standard gauge types in Finland.
Their descriptions are adapted from Huovila et
al. (1988), Førland et al. (1996) and Goodison
et al. (1998). The Wild precipitation gauge with
a Nipher wind shield was used, with minor variations, as the standard gauge in 1908–1981. The
cylindrical bucket of the Wild gauge is made
of a 0.5-mm-thick brass sheet. The orifice area
is 500 cm2 and its height is 150 cm above the
ground. Inside the bucket, there is a fixed conical
partition with a drain hole, some 4 mm in diameter. The wind shield is a truncated cone made of
galvanised steel.
In 1982, the Wild gauges were replaced by
the Tretjakov gauges. Their cylindrical vessel
is made of 0.5-mm-thick galvanised steel with
a varnished, grey outer surface. The orifice is
200 cm2 and it is placed at a height of 150 cm
above the ground. Inside the bucket there is a
cone-shaped partition with a large drainage hole.
To reduce evaporation during summer, the drain
hole is covered with a shield consisting of a
funnel with a hole 7 mm in diameter. The spout
of the vessel is about 215 mm from the bottom
of the bucket. The bucket is placed inside a
windshield made of 15 slats forming a cone, with
the upper edges of the slats bent outwards at an
angle of 70° to be horizontal. The slats are on a
level with the rim of the bucket.
A new standard Finnish bucket, H&H-90,
using the Tretjakov gauge wind shield, went into
operation at all manual observation stations on 1
4
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
found in Goodison et al. (1998). However, only
a fraction of the gauges have been operating concurrently; for instance in 2011 there were about
220 gauges measuring and reporting daily.
Form of precipitation
Fig. 1. Precipitation and regionally-representative wind
stations in 1961–2011.
January 1992. It is made of aluminium and covered inside with white PTFE and is painted white
on the outside. There is no spout in the bucket
and the height of the bucket is the same as that of
the Tretjakov bucket (40 cm).
In the latest years of the studied period,
an instrument based on the weighing principle, Vaisala All Weather Precipitation Gauge
VRG101 (Nitu and Wong 2010), was also used.
It has a similar windshield to that of the Tretjakov gauge and hence the same aerodynamic
correction method is assumed to be applicable
to it too.
Since 1961, WSFS has used precipitation
data from a total of 1051 gauges, of which 17,
6 and 9 are located in Sweden, Norway and
Russia, respectively (Fig. 1). The descriptions of
the gauges used Sweden and Norway can also be
Observations of the form of precipitation done by
the FMI were originally based on visual evaluation by an observer at the site. They were coded
with a three-digit integer, each digit describing
the form of precipitation recorded during 06:00–
12:00 UTC, 12:00–18:00 UTC and 18:00–06:00
UTC, respectively. The FMI categorised weather
into nine types (Table 1). In correction, some
precipitation types had to be combined: 0 and
6 were treated as dry weather, 1 was not taken
as rain and thus was not corrected, 2 and 7 were
treated as rain, and 3, 5, 8 and 9 as snow. Sleet
(type 4) was treated as 50% rain and 50% snow.
If, for example, the code for a particular day was
502 and the amount of precipitation 12 mm,
this was interpreted as solid precipitation of 4
mm recorded between 06:00 and 12:00 UTC,
dry weather between 12:00 and 18:00 UTC, and
liquid precipitation of 8 mm between 18:00 and
06:00 UTC. The amounts of precipitation were
assumed to be proportional to the precipitation
period lengths (in hours).
The FMI replaced the above-mentioned code
in 2007 with a relation developed by Koistinen
et al. (2004). It was derived using about 150 000
synoptic observations and discriminant analysis
and can be expressed as follows:
(1)*
where Plp is the probability for liquid precipitation, T is the temperature (°C), and H is the
humidity (%) at the height of 2 m. If Plp is
smaller than 0.2, precipitation is solid and if Plp
is greater than 0.8, precipitation is liquid. In the
case of 0.2 ≤ Plp ≤ 0.8 precipitation is treated as
sleet.
Wind
Wind speed and direction measurements are cru-
* On 29 September 2016 the equation was corrected in the following way: 2.2T (in the denominator) was replaced by 2.7T.
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
cial for the aerodynamic correction, together with
the exposure data of gauge sites. They are measured every third hour, being the most frequent
observation needed for the correction procedure.
Therefore, also correction was carried out using
the same frequency by first adjusting precipitation, form of precipitation and temperature observations to the same periods. Several types of
anemometers have been used since the beginning
of the 1960s. Some of them provided only a rough
estimate and the location of some devices was
not reliable enough. In this study, only regionally
representative wind speed and direction measurement stations defined by the FMI (see Fig. 1) were
used because of their reliability and also because
the aerodynamic correction method was developed for their data.
Correction method of precipitation
measurements
General
Because of the operational purpose of use, the
correction of the precipitation measurements
was carried out as a simplified version of the
Finnish exposure method developed by Solantie
(1986) based on the theory of Korhonen (1942).
Sarkanen (1989) and Førland et al. (1996) also
presented this approach. The overall correction
scheme consists of three components: (1) aerodynamic, (2) wetting and (3) evaporation correction. The common expression for corrected
precipitation Pc for the Nordic countries can be
written as (Førland et al. 1996):
Pc = (1 + A) ¥ (Pm + ΔPw + ΔPe),
Metadata
In addition to the hydro-meteorological data, the
metadata of the stations is needed for the correction of the precipitation observations. Besides
the gauge type information, the registration of
the height angle is crucial for the correction
procedure in eight directions (45° sectors) from
the gauge. The FMI has collected this information since 1961, dating the information every
two or three years if necessary, but it is by no
means comprehensive in terms of the number
of gauges and time coverage, thus affecting the
application of the method. Moreover, the height
of the anemometer is needed in the case of missing exposure information when the wind speed
during the storm is calculated.
5
(2)
where 1 + A is the aerodynamic correction, Pm
is the measured precipitation, ΔPw is the wetting
correction and ΔPe is the evaporation correction.
These terms are explained in detail next.
Aerodynamic correction
Typically, a freely-exposed precipitation gauge
systematically distorts the wind field, causing
the wind speed to increase above the gauge
orifice and forcing the development of eddies in
and around the gauge. Therefore, smaller liquid
and solid precipitation particles are prevented
from entering the gauge. Wind speed can be
considerably greater than the velocity of the fall-
Table 1. Codes for forms of precipitations used by the Finnish Meteorological Institute (FMI).
CodeExplanation
Weather
0
1
2
3
4
5
6
7
8
9
haze, smoke, snow drift, glaze, no precipitation
surface fog, mist, frost, rime, fog, dew
drizzle, rain, shower, freezing drizzle, freezing rain
small hail, frozen hail
sleet, sleet shower, ice pellets, rain with snow
snow, snow shower, snow pellets, ice needles
thunder without rain
thunder and rain
thunder and hail
thunder and snow or sleet
dry
humid
rain
hail
sleet
snow
thunder without rain
thunder and rain
thunder and hail
thunder and snow or sleet
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
6
where kwe is the exposure factor, kwa is the areal
wind coefficient (= 1 in this application) and kwi is
the actual wind coefficient. The factor kc depends
on the precipitation and gauge type (Table 2).
In the wind correction, kwe is defined as
100
80
a
60
40
a = 100exp(–b2/152.7)
20
0
0
5
10
15
b
20
25
30
Fig. 2. Instantaneous exposure coefficient a as a function of the height angle b (°) of the objects surrounding
the gauge (adapted from Sarkanen 1989).
ing particles, particularly snow, which results in
slow-falling particles being carried beyond the
lee side of the gauge orifice. Hence, the amount
of observed precipitation is generally smaller
than the non-biased precipitation. The loss of the
measured precipitation increases with the proportion of small drops and light snow particles
and with increasing wind speed (Sevruk 1989b).
The aerodynamic correction approach applied
in this study is called the exposure method and it
is developed particularly for conditions in which
wind speed observations at the gauge are not
available, but the weather data from the nearest
synoptic station can be used instead (Solantie
1986). The aerodynamic correction factor (1 +
A) depends on the wind speed and direction,
precipitation type, falling speed of the drops and
gauge type. The term A is called the relative aerodynamic correction and it consists of the wind
correction kw and the basic part kc:
A = kwkc = (kwekwakwi),
(3)
Table 2. Basic factor kc (adapted from Førland et al.
1996).
Precipitation
Temp.
Basic factor kc
type(°C)
Wild
Tretjakov and
H&H-90
Drizzle
– 0.340.05
Rain
– 0.010.01
Sleet
T ≥ 0
0.13
0.05
Snow
–8 < T < 0
0.28
0.18
Snow
T ≤ –8
0.40
0.18
Shower
T ≤ 1
0.28
0.18
Shower
T > 1
0.01
0.01
kwe = a/ ,
(4)
where
is the average exposure coefficient
(= 35 in this application) and a is the instantaneous exposure coefficient. The latter coefficient
is determined according to the wind direction
during the storm observed at the nearest synoptic
station, which is representative for the particular
area and describes the wind shelter given by surrounding structures and trees to the gauge using
the height angle b
b = tan–1(hobj/dobj),
(5)
where hobj and dobj are the height and distance of
the sheltering object, respectively. When b has
been obtained, a can be determined (Fig. 2). The
relation is interpreted so that a = 0 indicates a
fully sheltered and a = 100 a fully open gauge. All
the main wind directions are treated separately.
The actual wind coefficient kwi is expressed as
kwi = w/ ,
(6)
where w is the wind speed during the storm and
is the long-term average wind speed calculated
over all three hour observations, both obtained
from the nearest synoptic station which is representative for the particular area.
If there is no exposure information for the rain
gauge for the given time, the exposure method
cannot be used and kwe = 1. This modification is
called the wind method, in which the instantaneous wind speed and average wind speed are
reduced. The reduction was carried out by the
power law method (e.g. Schwarz et al. 2009):
w = w1(z/z1)0.143,
(7)
where w1 and z1 are the reference velocity and
reference height, respectively, and z is the height
of the gauge orifice. The exponent 0.143, suggested by Schlichting (1968), is often cited in
engineering texts for neutrally stratified boundary
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
layers. This assumption is valid for long-lasting
precipitation events or for events with hard wind.
There may be some snowfall in stable conditions,
but then wind speed is so small that the reduction
error due to assumption can be neglected.
Wetting and evaporation correction
The wetting correction ΔPw in Eq. 2 accounts
for precipitation that has become attached to the
inside walls of the precipitation gauge but is not
included in the measured amount. It depends on
the material of the gauge and the form of precipitation and is inversely proportional to the orifice
area of the gauge. With the exception of 80%
higher amounts for rain and snow in case of the
Wild gauge (R. Solantie pers. comm.), the values
used in this application (Table 3) are taken from
Førland et al. (1996).
The evaporation correction ΔPe estimates
the precipitation that had evaporated from the
gauge before the measurement has been taken.
The values of the mean daily evaporation loss
(Table 4) recommended by Førland et al. (1996)
are added to the measured value according to
Eq. 2. The fairly high values for the Tretjakov
gauge in the spring are explained by the practice of using the gauge without a funnel during
the cold period of the year. It is inserted in the
summer, providing the collector with a good
shelter from evaporation. The correction is made
in periods of three hours when precipitation has
occurred, and these values are adjusted correspondingly.
7
results. Since a vast amount of very heterogeneous data was used and the end product enables wide-ranging usage, the aim was to give a
thorough view of it and different related aspects
of the correction process. Henceforth, the time
interval from 1961 to 1981 when the Wild type
gauges were used is called Period1, and the time
interval from 1982 to 2011 when the Tretjakov
and H&H-90 gauges were in operation is called
Period2. The correction for both periods was carried out using the same method but with different
parameter values.
A total of about 9 million daily observations
1961–2011 were included, among them about
4.4 million (49.3%) days with precipitation. The
proportions of the forms of precipitation were
32.9%, 18.1% and 49.0% for solid, mixed and
liquid, respectively.
Daily correction factor in different
intensity classes and precipitation forms
First, histograms of the unitless correction factor
for different precipitation intensity classes were
examined. Both total precipitation and different
forms were considered (Fig. 3). For the sake of
clarity, the distributions were cut at 4.0, even
though there were higher values, but it did not
change their shape. The distribution of the correction factor varied clearly depending on the
form of precipitation within each intensity class
excluding < 1 mm day–1 (Fig. 3e–h); the mode of
distribution was smaller and its occurrence more
Table 4. Evaporation correction amounts (mm per day).
Results
Month
The following analysis was carried out to evaluate the correction method and reliability of the
Table 3. Wetting correction amounts (mm/case).
Precipitation type
Wild
Tretjakov
Rain
Drizzle
Snow
Mixed
0.1260.140
0.0500.120
0.0540.090
0.0700.140
H&H-90
0.130
0.090
0.050
0.110
Wild TretjakovH&H-90
January 0.010.03
February 0.020.04
March
0.030.05
April
0.010.22
May
0.010.13
June
0.010.15
July
0.010.15
August
0.010.10
September0.010.05
October 0.010.03
November 0.000.03
December 0.000.03
0.03
0.04
0.06
0.20
0.04
0.05
0.05
0.05
0.04
0.03
0.03
0.03
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
8
a ]0, ∞[, all
50
30
20
20
20
10
10
10
0
0
0
2.0
3.0
4.0
1.0
e ]0, 1], all
50
2.0
3.0
%
40
30
%
40
30
4.0
10
0
1.0
f ]0, 1], liquid
50
2.0
3.0
4.0
1.0
g ]0, 1], mixed
50
40
30
30
20
20
10
10
10
0
0
0
50
2.0
3.0
4.0
1.0
i ]1, 5], all
50
2.0
3.0
%
40
30
%
40
30
20
4.0
50
3.0
4.0
k ]1, 5], mixed
50
20
10
0
0
0
1.0
50
2.0
3.0
4.0
1.0
m ]5, 10], all
50
2.0
3.0
%
30
%
40
30
%
40
30
10
4.0
50
3.0
4.0
o ]5, 10], mixed
50
20
10
0
0
0
1.0
2.0
3.0
4.0
1.0
q ]10, ∞[, all
2.0
3.0
%
30
%
40
30
%
40
30
10
4.0
3.0
4.0
s ]10, ∞[, mixed
40
30
30
%
40
30
%
40
30
%
40
20
10
10
10
0
0
0
0
4.0
1.0
2.0
3.0
4.0
1.0
2.0
3.0
4.0
3.0
4.0
20
10
3.0
2.0
t ]10, ∞[, solid
50
2.0
p ]5, 10], solid
1.0
50
1.0
4.0
0
2.0
50
20
3.0
20
50
20
2.0
10
1.0
r ]10,∞[, liquid
l ]1, 5], solid
1.0
40
10
4.0
0
2.0
30
20
3.0
20
40
20
2.0
10
1.0
n ]5, 10], liquid
h ]0, 1], solid
1.0
40
10
4.0
0
2.0
30
20
3.0
20
40
20
2.0
10
1.0
j ]1, 5], liquid
d ]0, ∞[, solid
20
40
%
%
c ]0, ∞[, mixed
40
1.0
%
50
30
50
%
b ]0, ∞[, liquid
40
1.0
%
50
%
%
50
1.0
2.0
3.0
4.0
Fig. 3. Histograms of correction factor in different intensity intervals (in rows) and form classes (in columns) of
observed precipitation (mm day–1).
frequent for liquid than for solid precipitation.
On the other hand, the liquid form got high correction factor values with very small precipitation (Fig. 3f). The distribution of the correction
of the solid precipitation did not vary considerably with intensity (Fig. 3h, l, p and t). The correction factor of liquid precipitation was virtu-
ally 1.0–1.05 for higher intensities (> 5.0 mm)
(Figs. 3n and r).
Next, the 10th, 25th, 50th, 75th and 90th
percentiles of the correction factor for different
intervals and form classes of observed precipitation were studied (Table 5). The medians (50th
percentiles) of all precipitations in a way sum-
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
marise the correction on daily basis; the corrections for all forms, liquid, mixed and solid
precipitation in general were 1.21, 1.10, 1.21 and
1.39, respectively. Also on a general basis, the
higher the percentile, the greater the difference
of the correction factor between the forms. In
more detail, in each interval and percentile, the
correction factor of the solid form was greater
than the one of the mixed form, which was
greater than the one of liquid form, with the
exception of the precipitation interval ]0, 1] and
higher percentiles. These cases occurred during
spring and summer and were caused by the high
proportion of evaporation correction.
Amounts of correction components
The overall average amount of correction was
79.9 mm per year, equivalent to 11.9% of the
total corrected precipitation, and the reciprocal
proportions were 8.6%, 22.8% and 68.6% for
evaporation, wetting and aerodynamic corrections, respectively (Table 6). The aerodynamic
correction was clearly greater and the evap-
9
oration correction smaller in Period1 than in
Period2. Also the total amount of correction was
greater in Period1. The proportion of aerodynamic correction of the total correction (83.3%)
was remarkably high when solid was the most
typical from of precipitation (December–March).
It was clearly greater in Period1 (89.1%) than in
Period2 (77.1%). In May–September, the corresponding proportions were 38.8% and 23.3%.
In April–August, when evaporation correction
has a more notable role, the different correction
types were almost equal in Period2. In Period1,
the role of evaporation was minor.
Obviously, there was a great variation in the
monthly mean aerodynamic correction (Fig 4).
It can be almost as large as the mean in winter
months. In December–March, the amounts of
the aerodynamic correction were 8.6–12.0 mm
and 5.6–7.9 mm for Period1 and Period2, respectively. The average total amounts of the correction for the same months were 10.0–13.2 mm and
7.6–9.9 mm for Period1 and Period2, respectively.
The average aerodynamic correction amounts in
May–September were 0.7–1.8 mm and 0.7–1.1
mm for Period1 and Period2, respectively.
Table 5. Percentiles of correction factor in different intensity intervals and form classes of observed precipitation
(mm day–1).
Precipitation
Form of
Percentiles of correction factor
intensityprecipitation
(mm day–1)
10th25th50th70th90th
]0, ∞[
]0, 1]
]1, 5]
]5, 10]
]10, ∞[
all
liquid
mixed
solid
all
liquid
mixed
solid
all
liquid
mixed
solid
all
liquid
mixed
solid
all
liquid
mixed
solid
1.035
1.075
1.209
1.467
1.905
1.0261.0431.0951.2921.830
1.0691.1191.2091.3751.740
1.1171.2271.3851.6432.040
1.2001.2961.4931.8702.410
1.1911.2721.4671.9002.560
1.2141.2941.4501.7902.360
1.2111.3381.5451.8782.333
1.0501.0731.1241.2391.416
1.0461.0581.0821.1161.151
1.0701.1081.1611.2331.327
1.0731.1641.2831.4361.642
1.0251.0321.0451.1381.317
1.0241.0291.0351.0431.053
1.0361.0691.1131.1921.288
1.0491.1561.2661.4101.601
1.013
1.018
1.025
1.043
1.216
1.0131.0171.0221.0281.035
1.0261.0581.1061.1911.288
1.0481.1691.2861.4201.602
1961–2011
Evaporation6.9 8.6 1.01.8 4.5 1.03.118.7
Wetting 18.222.8 2.74.912.3 2.68.751.9
Aerodynamic
54.868.6 8.2
33.383.317.84.929.4
Sum
79.9100.0 11.940.0100.0 21.416.7100.0
Period1
Evaporation1.9 2.1 0.30.9 1.9 0.50.7 4.6
Wetting 16.718.8 2.54.2 9.0 2.38.356.5
Aerodynamic
70.579.1 10.7
41.389.123.25.738.8
Sum
89.1100.0 13.646.4100.0 26.114.6100.0
Period2
Evaporation
10.614.8 1.52.5 7.1 1.34.927.0
Wetting 19.326.9 2.85.415.8 2.88.949.6
Aerodynamic
41.858.3 6.1
26.677.113.54.223.3
Sum
71.6100.0 10.534.5100.0 17.518.0100.0
1.5 6.4 30.8 2.1
2.7 8.1 38.9 2.7
1.3 6.4 30.4 2.1
5.5 20.9 100.0 7.0
0.2 0.6 3.5 0.2
2.7 7.3 39.5 2.6
1.910.5 57.0 3.8
4.8 18.5 100.0 6.7
1.0 4.0 19.8 1.4
2.7 7.8 38.8 2.7
1.6 8.3 41.4 2.9
5.3 20.1 100.0 6.9
Period January–December,December–March, May–September,
April–August,
liquid and solid precipitation
mainly solid precipitation
mainly liquid precipitation
large evaporation correction
Amount Percentage PercentageAmountPercentagePercentageAmount Percentage Percentage Amount PercentagePercentage
(mm) of correction of precip.
(mm) of correction of precip.
(mm) of correction of precip.
(mm) of correction of precip.
Table 6. Average corrections for evaporation, wetting and aerodynamic losses per year for mainly snow, rain and evaporation periods.
10
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
20
11
a
18
16
14
12
10
8
6
4
2
0
–2
20
b
18
16
14
12
10
8
6
4
2
0
–2
16
c
14
12
10
8
6
4
Fig. 4. Mean monthly
amounts (± SD, mm) of
correction components.
(a)
1961–2011,
(b)
Period1, and (c) Period2.
2
0
–2
Jan
Feb
Mar
Apr
May
evaporation
In April–August, the different amounts of
evaporation correction were emphasised. They
were 0.1 mm and 0.8–2.2 mm for Period1 and
Period2, respectively. Wetting correction varied
least through the months and between the periods,
the ranges being 0.9–2.0 mm and 1.1–2.0 mm for
Period1 and Period2, respectively (Fig. 4b and c).
Jun
wetting
Jul
Aug
Sep
Oct
Nov
Dec
aerodynamic
Effect of exposure information
To understand the effect of exposure information of the gauges on correction amounts, the
same set of data was corrected by both exposure
and wind methods. The data set consisted of
the period 1961–1999 and 75 gauges, including
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
12
15
15
a
10
%
%
10
b
5
5
0
0
0
15
200
400
600
Precipitation (mm)
800
0
15
c
800
d
%
10
%
10
200
400
600
Precipitation (mm)
5
5
0
0
0
200
400
600
Precipitation (mm)
800
0
the Wild, Tretjakov and H&H-90 types, with
reliable exposure information. In case of liquid
precipitation (Fig. 5a and c), the distributions of
the corrected yearly sums seem quite similar. The
means of liquid precipitation sums corrected by
exposure and wind methods were 433.5 mm (SD
= 115.0 mm) and 433.9 mm (SD = 115.2 mm),
respectively. When statistically tested (paired
t-test assuming unequal variances, e.g. Milton and
Arnold, 1987) they differed significantly (t2685 =
2.7605, p = 0.00581). In case of solid precipitation, the distributions differed more clearly (Fig.
5b and d). The means were 216.0 mm (71.4 mm)
200
400
600
Precipitation (mm)
800
Fig. 5. Histograms of
yearly precipitation sums
of 75 gauges in 1961–
1999: (a) liquid precipitation corrected by the wind
method, (b) solid precipitation corrected by the
wind method, (c) liquid
precipitation corrected
by the exposure method,
and (d) solid precipitation
corrected by the exposure
method.
and 220.6 (72.4 mm) for exposure and wind
methods, respectively. Paired t-test assuming unequal variances (t2685 = 6.7384, p = 1.952 ¥ 10–11)
also showed clear evidence of the difference.
Regional distributions of annual
precipitation sums
There are some characteristic places in Finland in terms of precipitation (Fig. 6a), which
should be mentioned when precipitation maps,
obtained by calculating areal precipitation to
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
a
Enontekiö Ma
a
th
n
bo
st
ro
O
Su
e
f
elk
ä
ä
elk
ns
e
om
c
Kainuu
ia
Gulf of Bothnia
ns
b
13
Lake
area
Tammela
Baltic
Sea
d
Gulf of Finland
Fig. 6. Spatial variation of annual mean precipitation (mm) in Finland: (a) regions, (b) observed, (c) total corrected,
(d) total corrected – observed, (e) corrected liquid, and (f) corrected solid. The areas in the background are the
main watershed boundaries.
a 1 ¥ 1 km grid by inverse distance squared
method using the three nearest point-wise observations (Fig. 6b–f), were analysed. Minimal spatial smoothing between grid cells was used to
study the effects of different gauges. All the distinguishable spots were checked. In some cases,
a gauge was located in such an open place that it
was exposed to winds and much higher aerodynamic correction was carried out in comparison
with the surrounding gauges. However, there
were also cases when measurements of a gauge
were found to be unreliable and were removed
from the data.
The general patterns of the observed (Fig. 6b)
and corrected (Fig. 6c) precipitation were increasing towards the east and south of Finland. This
was more emphasized in corrected amounts. Of
course, there were local anomalies from this pattern due to exposure and wind conditions. The
general pattern of difference between corrected
and observed precipitation (Fig. 6d) followed the
one of the solid precipitation (Fig. 6f). Liquid
precipitation (Fig. 6e) was greater than solid
precipitation. Moreover, the areal distribution
of precipitation was different in the winter and
summer halves of the year. In winter, the Scandi-
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
14
navian mountains dried out the precipitation areas
coming from west with the result of increasing
the precipitation in Finland about 5% per 100 km
towards the south-east of the country. Other areas
with high solid precipitation (300–350 mm) were
the Maanselkä area, the eastern ends of the Suomenselkä and particularly the Enontekiö area.
There were also local spots of high amounts of
solid precipitation outside these areas. On the
other hand, the coast of the Gulf of Bothnia, the
Ostrobothnia area between the coastline and Suomenselkä, and south–west Finland were the areas
with the least solid precipitation (150–200 mm).
Overall, the total regional yearly sum was
composed of liquid and solid totals. The maximum was located in the same place as the solid
maximum, and the minimum in the same place
as the liquid minimum. The features of the
above-mentioned distribution of solid and liquid
precipitation stand out better in the maps of total
sums, for instance the tendency of increasing
solid precipitation from north-west to south-east
and the higher amounts close to the watershed
divides. These phenomena were more emphasised with the corrected precipitation (Fig. 6c).
1982 was omitted since the replacement of the
Wild gauges by the Tretjakov gauges took place
during that time. Obviously, the variation of
the corrected total was greater than that of the
observed precipitation (Fig. 7a and b), particularly the large values are more frequent with
the corrected total. The differences of means
between the periods were statistically tested
using Student’s t-test for unequal variances (see
e.g. Milton and Arnold 1987) and were found
statistically highly significant which indicates
that the observed, corrected total and corrected
liquid precipitation were greater and corrected
solid precipitation smaller in Period2 (Table 7).
The statistics of linear trend analysis was
based on the least square method of slope fitting
(Table 8). In the whole period, the trends were
found in all parameters, three of them being positive and the one for the solid precipitation negative. In Period1, the trends in all parameters were
positive trends, whereas in Period2 the trend was
positive for liquid precipitation only and negative for both corrected total and corrected solid
precipitation, with observed precipitation having
no trend at all.
Long-term time series and linear trends
of annual sums
Validation using a hydrological model
Considering the effect of correction for all
gauges and the whole period, the total annual
precipitation increased 13.6% from 590.3 mm to
670.6 mm. The proportions for liquid and solid
precipitation were 63.6% and 36.4%, respectively (Table 7).
The long-term analysis of annual sums of
observed and corrected precipitation (total, solid
and liquid) was carried out separately for the
whole period, Period1 and Period2. The year
First, appropriate metrics for assessing the goodness of precipitation correction was defined.
A common metrics in hydrological modelling,
coefficient of determination r2 (also known as
Nash-Sutcliffe model efficiency), was calculated
for the observations and simulations of WSFS
discharge simulations. We found out that comparing r2 values from runs with uncorrected and
corrected precipitation data gave a clear indicator of the goodness of precipitation correction
in terms of water balance. To distinguish the
Table 7. Mean annual precipitation sums (mm) for 1961–2011, Period1 and Period2 and results of the t-test for
unequal variance comparing Period1 with Period2.
Observed
Corrected total
Corrected liquid
Corrected solid
1961–2011Period1 Period2
590.3
670.6
439.6
231.0
566.4
657.4
418.2
239.2
611.5
683.5
458.2
225.3
df
20 687.1
20 072.4
21 862.3
20 988.7
t
p
–29.0528
–15.0952
–27.9645
13.6527
< 2.2 ¥ 10–16
< 2.2 ¥ 10–16
< 2.2 ¥ 10–16
< 2.2 ¥ 10–16
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
1400
15
a
Precipitation (mm)
1200
1000
800
600
400
200
0
1400
b
Precipitation (mm)
1200
1000
800
600
400
200
0
1000
c
Precipitation (mm)
800
600
400
200
0
Precipitation (mm)
1000
d
800
600
400
200
0
1961
1966
1971
1976
1981
1986
1991
1996
2001
2006
2011
Fig. 7. Time series of annual precipitation (mm) distribution: (a) observed, (b) total corrected, (c) corrected liquid,
and (d) corrected solid. Dark grey boxes indicate Period1, light grey boxes Period2, and the white box the gauge
exchange year 1982. The horizontal line within the box indicates the median; the bottom and top of the box the 25th
and 75th percentiles, respectively; the upper whisker shows 1.5 times the interquantile range or maximum value,
whichever is the smaller, and the lower whisker 1.5 times the interquantile or minimum value, whichever is the
greater; circles are outliers.
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
–0.1420.124 –1.15 0.2521
–1.056
0.134
–7.87
< 0.0001
0.280
0.122 2.300.0213
–1.336
0.081
–16.58
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
13.43
13.24
8.95
11.29
0.231
0.264
0.204
0.148
3.106
3.501
1.827
1.674
< 0.0001
< 0.0001
< 0.0001
< 0.0001
28.24
13.59
27.12
–15.62
0.060
0.067
0.056
0.039
1.687
0.905
1.509
–0.604
Observed
Corrected
Liquid
Solid
tp
Parameter1961–2011
Period1
Period2
EstimateSE
tp
EstimateSE
tp
EstimateSE
Table 8. Statistics of long-term analysis of annual precipitation. Estimates set in boldface, italic and underlined indicate positive, negative and no trend, respectively.
16
r2 difference due to uncorrected and corrected
precipitation, r2 values were calculated only for
the period 1 March–20 June in 26 years determined by the WSFS calibration procedure. This
is because the correction for solid precipitation
is much greater than for liquid precipitation and
cumulative precipitation during winter strongly
affects spring flood discharge simulation and
therefore differences in r2 are greater in spring
than in other seasons.
The r2 metrics simulations were run with precipitation input produced by four methods: (1)
traditional correction, (2) daily correction, (3)
observed precipitation, and (4) daily correction
with fitted coefficients. Traditional correction
refers to method used in WSFS in which fitted
areal coefficients for both liquid and solid precipitation are determined and division to solid
and liquid precipitation was done with fitted
areal temperature limit values. Coefficients and
limit values were fitted to minimize the difference of simulated and observed discharge,
which will give the traditional method an edge
over daily correction method when compared
with r2 metrics. Daily correction method is the
method described in this article. When using
measured precipitation, the form of precipitation
is determined by observations and the precipitation volumes are not corrected but plain gauge
observations are used. The method of daily correction with fitted coefficients is a combination
of daily correction method and traditional correction method: fitted areal coefficients for both
liquid and solid precipitation are applied to precipitation corrected by the daily method.
There was no substantial difference between
the daily and traditional correction methods (see
Fig. 8a). The histogram is slightly tilted in favour
of the traditional method which, however, is
expected due to fitted coefficients and temperature limit values. Using daily corrected precipitation results in greater r2 in most discharge stations than using plain observations (Fig. 8b). The
r2 differences between the traditional correction
and the daily correction with fitted coefficients
are depicted in Fig 8c. Superiority of either
method is not obvious, but the results resemble
the ones presented in Fig. 8a. However, it must
be noted that using daily corrected precipitation with fitted coefficients instead of traditional
Fig. 8. Histograms of
differences in goodness
of fit (r 2) of hydrological
model simulations with
precipitation inputs from:
(a) daily correction – traditional correction, (b) daily
correction – observed
precipitation, and (c) daily
correction with fitted coefficients – traditional correction.
Frequency of occurrence in the data
Frequency of occurrence in the data
Frequency of occurrence in the data
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
600
17
a
500
400
300
200
100
0
600
–0.5
–0.4
–0.3
–0.2
–0.1
0
0.1
0.2
0.3
0.4
0.5
–0.4
–0.3
–0.2
–0.1
0
0.1
0.2
0.3
0.4
0.5
–0.4
–0.3
–0.2
–0.1
0
0.1
0.2
0.3
0.4
0.5
b
500
400
300
200
100
0
600
–0.5
c
500
400
300
200
100
0
–0.5
method increased the goodness of fit more than
0.10 for some discharge stations. Comparison
of Fig. 8a and c indicates that using areal fitted
coefficients with daily correction is beneficial.
To validate the correction factor of daily correction method, we examined the respectively
fitted areal coefficients, which alter the precipitation input to better match the water balance
based on discharge observations (Fig. 9). Coefficients below 1 indicate that the daily correction
method is correcting the volume of precipitation
too much and coefficients above 1 indicate a too
small daily correction.
Difference in goodness of fit
Discussion
The focus of this study, outlined on the basis
of operational hydrological modelling in which
every observation is valuable if it is reliable, was
to check the correction procedure and estimate the
amount of correction in total precipitation and in
liquid and solid forms. The correction method was
selected so that it suits well the Finnish conditions
and uses standard hydro-meteorological observations readily available for the whole study period.
Data were heterogeneous due to changes in gauge
network, a number of gauge and instrument types,
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
18
1000
a
Frequency
800
600
400
200
0
0.90 0.925 0.95 0.975 1.00 1.025 1.05 1.075 1.10
1000
b
Frequency
800
600
400
200
0
0.90 0.925 0.95 0.975 1.00 1.025 1.05 1.075 1.10
Areal coefficients
Fig. 9. Histograms of areal coefficients for (a) solid and
(b) liquid precipitation fitted with the corrected precipitation to match the water balance based on discharge
observations.
countries and practices, exposure, geographical
and climatic conditions, etc. The principle of synchronising precipitation, wind and form of precipitation observations suits well with stable weather
conditions but can bring on errors with unstable
conditions such as cold fronts and showers. Moreover, daily mean temperature value is too rough
for many weather situations since temperature can
vary significantly during a 24-hour period. Using
only reliable, regionally representative wind stations guarantees that the general weather type is
well described and thus supports areal precipitation calculations, but diminish sensitivity to adapt
to spatial variation. Allerup et al. (2000) stated
that in Danish conditions wind observations can
be extrapolated from remote sites not farther away
than approximately 50 km. It is unknown how
this applies to Finnish conditions but we estimated that approximately in 75% of all cases this
instruction could be complied.
Wagner (2009) stated, based on literature
review and questionnaire, that most precipita-
tion data are still not systematically corrected.
There are also very few published studies about
operational and automatic precipitation corrections concerning all observations or observations
from many decades with large spatial dimensions of a certain country or weather services
institute. Most often, a limited number of gauges
and/or homogenous conditions are selected, e.g.
12 stations in Greenland 1994–1997 by Yang et
al. (1999), a few gauges in Norwegian Arctic
by Førland and Hanssen-Bauer (2000), drifting
stations in the Russian North Pole 1950–1990
by Bogdanova et al. (2002), five stations in highlatitude regions from three winters by Sugiura et
al. (2006). The lack of operational corrections
is probably due to considerable amount of data
processing needed. Moreover, conditions can
vary a lot and therefore a suitable method is not
easily found, particularly for solid precipitation. Some of the very few studies found in the
literature regarding long-term nationwide precipitation correction were the ones obtained by
Lapin and Šamaj (1991) for a network of about
700 gauges in Slovakia 1981–1988 and Ye et al.
(2004) for 710 stations in China 1951–1998.
Most studies comparing the corrected
and observed precipitation are carried out for
monthly or yearly sums even though in operational hydrology the usage of the measurements
takes place often on a daily level. In one of the
few assessments of daily amounts, Yang et al.
(1998) stated that small absolute differences of
measurements can result in large variations in
catch ratios in daily precipitation. Therefore, in
their analysis they only accounted for amounts
greater than 3.0 mm per day, and concluded that
the undercatch was always greater for solid than
for liquid or mixed precipitation. Legates et al.
(2005) found out that Arctic precipitation correction carried out using daily data are significantly
more accurate than those based on monthly data
because the daily meteorological data are more
representative of the actual conditions when precipitation occurred. The results of this study
showed that liquid form gets higher correction
factor values more often. This obviously is due
to the greater evaporation correction during the
summer months. The relative effect of evaporation correction, as well as the one of the wetting correction, is particularly substantial with
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
low precipitation intensities (< 0.2 mm h–1). The
correction of the solid precipitation occurs in a
different way; it does not vary significantly with
intensity because of its strong dependence on the
aerodynamic correction. Instead, in cold weathers (T ≤ –8 °C) it tends to get greater correction
due to light snowflake.
Different correction methods, observation
instruments and weather and climate conditions
can produce remarkably different results for the
corrected amount of precipitation and therefore
comparing results obtained in this study with
those of others’ can only be done on a very general level. For example, Yang (1999) obtained
256 mm and Bogdanova et al. (2002) 165 mm
for annual average corrected precipitation of
Russian North Pole drifting stations in 1957–
1990. At Alaska, correcting the Canadian Nipher
gauge data Benson (1982) reported a correction
factor, without considering wetting loss and trace
amounts, of 1.1 and 3.5 for liquid and solid precipitation, respectively. Legates and Willmott
(1990) presented monthly based climatological
correction factors and, for instance, for Finland
on January 1999 (virtually solid precipitation)
they were greater than 1.5. Fuchs et al. (2001)
showed similar values, estimated by an eventbased method, for the south coast of Finland
and Enontekiö area in the north but somewhat
smaller values (1.1–1.5) for other areas. Yang et
al. (1998), correcting wetting loss before dealing with the wind-induced errors, indicated an
overall correction factor of 1.2 and 1.9 for liquid
and solid precipitation, respectively. Metcalfe
et al. (1994) corrected the snow data and the
results indicated that the total corrected annual
precipitation was 50%–100% greater than the
gauge-measured yearly total. As for the Tretjakov gauges, also Sevruk (1982), Groisman et al.
(1991) and Yang et al. (1995) demonstrated that
the correction for snowfall is greater than that for
rainfall. Førland et al. (1996) found a catch ratios
of 0.95 and 0.7 for liquid and for solid precipitation, respectively, in their study with Nordic
gauges. Førland and Hanssen-Bauer (2000)
found a ratio between corrected and measured
precipitation of 1.26 and 1.70 for the summer
and winter seasons, respectively, in Norwegian
Arctic. Yang and Ohata (2001) applied several corrections methods for Tretjakov gauges in
19
Siberian regions in 1986–1992. Wind-induced
undercatch was found to be the greatest error and
the annual precipitation was found to increase
10%–65% due to correction. Adam and Lettenmaier (2003) showed 18.3% increase in mean
annual precipitation for latitudes 60°–70°N in
Eurasia using their systematic bias adjustments.
Zhang et al. (2004) concluded that the corrected
annual precipitation was 17%–42% higher than
those measured and corrections were found to be
higher in winter than in summer time in Mongolia. The wind loss was 5%–30% and wetting loss
3%–9% of the total correction. The evaporation
loss was 4%–11%, which is slightly higher than
2%–8% found by Groisman et al. (1991) for
Siberian conditions. Sugiura et al. (2006) stated
53.9% catch ratio for solid precipitation.
Also this study showed an obvious difference between liquid and solid precipitation forms
in monthly and yearly sums. The correction of
solid precipitation was particularly emphasized in
cold weathers and strong winds from unsheltered
directions. The proportions between corrected and
observed amounts, as well as different components, were well within the ranges reported in the
literature. The aerodynamic correction being the
biggest, according to both this and other studies,
emphasizes the influence of the method used. In
practice, it is difficult to avoid placing precipitation gauges in locations with surrounding obstacles such as bushes, trees and buildings. At the
same time in operational hydrology, there often is
a shortage of information. Therefore, the exposure
method presented and applied in this study can
compensate many of the deficiencies or rectify
errors and provide reliable precipitation information. There are only a few surveys (e.g. Solantie
1986, Førland et al. 1996) reporting usage of
exposure information of the gauges when correcting precipitation observations. In this study,
utilization of this information was found important and caused statistically significant difference
in results on a general level. The differences were
analysed in a long period (1961–1999) using 75
gauges without paying attention to the direction
of the wind during the precipitation event so it can
be assumed that there were a lot of occasions both
in sheltered and unsheltered wind conditions.
The general patterns of the precipitation maps
were found to be in strong correspondence with
20
the ones presented in Solantie (1987). The precipitation amounts were higher in summer than
in winter due to the showery nature of the rainfall
and higher moisture content of the atmosphere
then. It is also possible that some heavy snow or
sleet showers occurring near 0° were interpreted
as liquid precipitation. Anyway, the influence of
this phenomenon should be marginal not causing
bias in general patterns. The absolute amount of
correction increasing towards the east of Finland, particularly in the southern parts, can be
explained by the fact that the observed precipitation and thereafter the corrected precipitation is
increasing even more strongly. The orographic
conditions showed an important role during
winter months since the rain clouds were low,
stated also by Solantie (1987) and Solantie and
Pirinen (2006). This explains the higher solid
precipitation amounts in the areas of watershed
divides, e.g. in the Salpausselkä area and in the
higher ground of Tammela. On the other hand,
during the summer half of the year, orography
has virtually no effect on the regional distribution of precipitation, since the rainclouds are so
high. However, the heat characteristics of land
and water are significant, particularly in inland
areas. In the afternoons, there is a tendency for
rising currents above warm land surfaces, leading to showers and therefore liquid precipitation amounts are typically higher in inland than
in other areas (Solantie 1987). However, due
to the long-term averaging and because all the
precipitation does not have this nature, the scattered effect of showers is not obvious in the lake
area, but can easily be seen in the vicinity of the
southern coast line. During the summer, the sum
effect of central Finland, being on the route of
lows moving from west to east and surrounded
by sea and lakes, causes maximum liquid precipitation in Suomenselkä, the lake area and
Kainuu. The similarity between the difference
of corrected and observed precipitation and the
corrected solid precipitation patterns was due to
the fact that the correction of solid precipitation
was greater. Corresponding regional patterns can
be recognized in many snow related records,
like solid form precipitation as percentage of
total precipitation, snow depth with a rare occurrence (Solantie 1987), winter month precipitation amounts (Pirinen et al. 2012), snow zones
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
(Solantie et al. 1996) and also from some temperature related records, such as the annual range
of monthly mean temperatures (Helminen 1987).
Instead, similarity with regional wind patterns
is not obvious, apart from the SE corner of Finland which is prone to winds from the direction
between west and south (Helminen et al. 1987).
The effect of winds depends on the exposure
conditions of the gauge, for instance even strong
winds can results in minuscule correction if the
gauge is sheltered from the direction of wind.
The effect of wind can also be difficult to distinguish from the general climate patterns because
the wind direction during the precipitation event
(measured every third hour) can vary and gauges
may have different exposure conditions in each
direction.
Even though statistically significant trends
were found in terms of different periods and
precipitation forms, a stand, based solely on this
study, cannot be taken on whether they are due
to changes in gauge types, gauge network, correction method, climate, etc., or a combination
of these. The results are also not representative
averages for Finland since the spatial distribution of gauges is not even throughout the country
(Fig. 1). When studying changing climate in
particular, a representative group of gauges with
carefully analysed metadata should be selected
and used only in the analysis, as was done by
Pirinen et al. (2012).
Saltikoff et al. (2015) compared radar based
estimates of solid precipitation with the corrected solid precipitations of this study. They
found that their average ratio was 1.59 for the
entire data set. The ratio depended on the precipitation intensity: the weaker the intensity was,
the greater the ratio. This was most common
for precipitation below 2 mm per day. On the
other hand, in moderate and high-intensity precipitation, they found cases in which corrected
precipitations were greater than the radar-based
estimates. The ratio varied with the distance to
the radar so that at shorter ranges radar precipitation was typically greater than the corrected
precipitation but beyond 150 km the radar gave
increasingly smaller values.
In the assessments with the watershed scale
hydrological model, we found that using the
corrected precipitation instead of plain observed
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
precipitation improved the goodness of fit of
the model (r2), and using the fitted coefficients
with daily correction improved it even more.
However, using either the traditional method or
the correction method of this study with fitted
coefficients did not make significant difference.
The fact that no clear distinction in r2 values
was found means that it is worthwhile to use the
corrected precipitations with areal fitted coefficients as the input for the hydrological model,
because, in principle, they are more realistic,
taking into account the ever-changing weather
circumstances at gauges. Our findings are consistent with the ones of Stisen et al. (2012). They
applied two precipitation correction methods to
a historic 20 year record, and evaluated the
resulting precipitation data sets through a comprehensive water resources model in Denmark.
They found that simulated stream discharge was
improved significantly and optimized model
parameters were much more physically plausible
using the other correction method. However,
our results are only preliminary since a more
profound research with WSFS could not easily
be implemented herein. This analysis supports
a comprehensive future research with new calibrations of the modelling system using different
precipitation data sets made explicitly comparable. The amount of liquid precipitation correction was considered approximately correct and
the amount of solid precipitation correction a bit
too large. The results indicated that the coefficients for solid precipitation might be fitted even
below the limitation 0.9–1.1.
Conclusions
Aerodynamic, wetting and evaporation correction was carried out for daily precipitation observed in Finland and its cross-boundary catchments in 1961–2011. The correction
method was found easy to apply to Finnish
conditions. The most time-consuming part was
obtaining and combining data from different
sources and replacing missing data.
In comparison between corrected and
observed precipitation, the proportions and
amounts were found to be realistic and in accordance with those in the literature. The mean yearly
21
precipitation sums of all the stations and the
whole period were 590.3 mm and 670.6 mm for
observed and corrected precipitation, respectively,
which is equivalent to a 13.6% increase. The general pattern of spatial variation of total corrected
precipitation and its different forms corresponded
to the prevailing knowledge and former research.
The distribution of the correction factor
varies so that the mode was smaller and its
occurrence more frequent for liquid than for
solid precipitation in the intensity intervals with
the exception of the lowest one (< 1 mm day–1).
This was due to the greater aerodynamic correction of the solid form. With low intensities
during the summer months, the liquid form had a
more frequently higher correction factor because
of the greater evaporation correction. The variation of the correction factor was virtually negligible with liquid precipitation greater than 5
mm day–1. The median correction factor on daily
basis was 1.21.
In the comparison of different components,
the average overall proportions were 8.6, 22.8
and 68.6% for evaporation, wetting and aerodynamic corrections, correspondingly. In the
case of solid precipitation, the proportions of
the aerodynamic correction were about 89.1%
and 77.1% for Period1 (Wild gauges 1961–
1981) and Period2 (Tretjakov 1982–1991 and
H&H-90 1992–2011), correspondingly. In the
case of liquid precipitation, these proportions
were 38.8% and 23.3%.
In the analysis of the long-term annual precipitation sums, the observed, corrected total
and corrected liquid precipitation were found
to be greater and corrected solid precipitation
lesser in Period2 than in Period1. For the entire
studied period, the linear trend analysis indicated
increasing amounts of observed, corrected total
and liquid precipitation and decreasing amounts
of corrected solid precipitation. In Period1, all
the trends were positive but in Period2 the trends
were mixed: positive for liquid precipitation and
negative for both corrected total and corrected
solid precipitation. It cannot be rejected that
these trends are real, but they remain obscure for
the time being because the vast amount of data
used included a lot of heterogeneity.
Although conclusions cannot be drawn on
climatic issues, this study provides a large, care-
22
fully checked data source to be used for operational hydrological modelling, as originally
intended, and also for hydrological and climate
research after appropriate data selection for the
purpose. According to our literature review,
studies on systematic, operational correction of
all daily precipitation observations during fifty
years are very rare. The correction procedure
was developed so that it can be executed automatically when a new observation is received.
The procedure utilizes available information in a
diverse and effective way using alternative methods or data replacement schemes. We showed
that accounting for exposure information of the
gauges has a statistically significant decreasing effect on correction amounts. Our method
evaluation is not only based on the literature
review but also on comparisons with radar and
hydrological model based estimates. However,
the conclusions drawn are divergent for solid
precipitation; the former comparison indicates
underestimation and the latter overestimation
on the average level when using our procedure.
Liquid precipitation correction is instead on a
correct level indicated by the hydrological model
comparison. Using corrected precipitation as
input in hydrological model simulations seems
favourable but a more comprehensive study on
this topic is recommended.
Acknowledgements: The authors are particularly grateful
to Mr. Pauli Rissanen (Finnish Meteorological Institute,
retired) for his ideas and guidance in applying the presented
procedure and for providing the data, to Dr. Reijo Solantie
(Finnish Meteorological Institute, retired) and Dr. Bertel
Vehviläinen) for their valuable comments on this study, to
Mr. Ari Koistinen (Finnish Environment Institute) for the
rewarding discussions on the hydrological model and to Mr.
Kalle Sippel and Mr. Juha Markkula (Finnish Environment
Institute) for their help in producing the figures.
References
Adam J.C. & Lettenmaier D.P. 2003. Adjustment of global
gridded precipitation for systematic bias. J. Geophys.
Res. 108(D9), 4257, doi:10.1029/2002JD002499.
Allerup P. & Madsen H. 1980. Accuracy of point precipitation measurements. Nord. Hydrol. 11: 57–70.
Allerup P., Madsen H. & Vejen F. 1997. A comprehensive
model for correcting point precipitation. Nord. Hydrol.
28: 1–20.
Allerup P., Madsen H. & Vejen F. 2000. Correction of pre-
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
cipitation based on off-site weather information. Atmos.
Res. 53: 231–250.
Bell T.L. & Kundu P.K. 2003. Comparing satellite rainfall
estimates with rain gauge data: Optimal strategies suggested by a spectral model. J. Geophys. Res. 108(D3),
4121, doi:10.1029/2002JD002641.
Benson C.S. 1982. Reassessment of winter precipitation
on Alaska’s Arctic slope and measurement on the flux
of wind blown snow. Report UAG R-288, Geophysical
Institute, University of Alaska, Fairbanks.
Bogdanova E.F., Ilyin B.M. & Dragomilova I.V. 2002. Application of a comprehensive bias-correction model to
precipitation measured at Russian North Pole drifting
stations. J. Hydrometeorol. 3: 700–713.
Chvíla B., Sevruk B. & Ondrás M. 2005. The wind-induced
loss of thunderstorm precipitation measurements. Atmos.
Res. 77: 29–38.
Collier C.G. 1986a. Accuracy of rainfall estimates by radar,
part I: calibration by telemetering raingauges. J. Hydrol.
83: 207–223.
Collier C.G. 1986b. Accuracy of rainfall estimates by radar,
part II: comparison with raingauge networks. J. Hydrol.
83: 225–235.
Dahlström B. 1970a. A general classification of error sources
in raingauging and some applications. In: Nordisk Hydrologisk Konferens, Stockholm 27–29 Aug. 1970, vol. 2,
Nordisk Hydrologisk Förening, Lund, pp. 57–70.
Dahlström B. 1970b. Automatic error checking and correction of rainfall data. In: Nordisk Hydrologisk Konferens,
Stockholm 27–29 Aug. 1970, vol. 2, Nordisk Hydrologisk Förening, Lund, pp. 79–87.
Dahlström B., Aune B., Førland E., Madsen H., Perälä J. &
Solantie R. 1986. The improvement of point precipitation
data on an operational basis. Description of a general
method for point precipitation correction with regard
to application in practice. NHP Report no. 17, NHP
Working Group on Precipitation, Nordic Hydrologic
Programme (NHP), SMHI, Norrköping, Sweden.
Finland’s Fifth National Communication under the United
Nations Framework Convention on Climate Change
2009. Ministry of the Environment and Statistics Finland, Helsinki, Finland.
Førland E.J. & Hanssen-Bauer I. 2000. Increased precipitation in the Norwegian Arctic: true or false? Clim.
Change 46: 485–509.
Fuchs T., Rapp J., Rubel F. & Rudolf B. 2001. Correction
of synoptic precipitation observations due to systematic
measuring errors with special regard to precipitation
phases. Phys. Chem. Earth B 26: 689–693.
Førland E.J., Allerup P., Dahlström B., Elomaa E., Jónsson
T., Madsen H., Perälä J., Rissanen P., Vedin H. & Vejen
F. 1996. Manual for operational correction of Nordic
precipitation data. Nordic working group on precipitation. Report no. 24/96, Norwegian Meteorological Institute, Oslo.
Gjertsen U. & Dahl J.I. 2001. Development of a gauge
adjustment method for correction of radar precipitation
estimates. DNMI Research Report no. 128.
Goodison B.E., Sevruk B. & Klemm S. 1989. WMO solid
precipitation measurement intercomparison: objectives,
BOREAL ENV. RES. Vol. 21 • Correction of daily precipitation measurements
methodology, analysis. In: Delleur J.W. (ed.), Atmospheric deposition, IAHS Publication 179, Baltimore,
MD, pp. 57–64.
Goodison B.E., Louie P.Y.T. & Yang D. 1998. WMO solid
precipitation measurement intercomparison. Final
report. Instruments and observing methods, report no.
67. WMO/TD No. 872, World Meteorological Organization, Geneva, Switzerland.
Groisman P.Ya., Koknaeva V.V., Belokrylova T.A. & Karl
T.R. 1991. Overcoming biases of precipitation measurement: a history of the USSR experience. Bull. Am. Meteorol. Soc. 72: 1725–1732.
Helminen V.A. 1987. Temperature conditions. In: Alalammi
P. (ed.), Atlas of Finland. Folio 131, Climate, National
Board of Survey and Geographical Society of Finland,
Helsinki, Finland, pp. 2–5.
Helminen V.A., Sarkkula S. & Solantie R. 1987. Atmospheric pressure and winds. In: Alalammi P. (ed.), Atlas of
Finland, Folio 131, Climate, National Board of Survey
and Geographical Society of Finland, Helsinki, Finland,
pp. 5–6.
Huovila S., Elomaa E., Leminen K., Tammelin B. & Tuominen A. 1988. Comparison of snow gauges used in Nordic
countries — Contribution of Finland to WMO solid precipitation measurement intercomparison. Part I: System
description. Meteorological publications no. 9, Finnish
Meteorological Institute, Helsinki, Finland.
Huttunen I., Huttunen M., Piirainen V., Korppoo M., Lepistö
A., Räike A., Tattari S. & Vehviläinen B. 2016. A
national scale nutrient loading model for Finnish watersheds — VEMALA. Environmental Modeling and
Assessment 21: 83–109.
Koistinen J., Michelson D.B., Hohti H. & Peura M. 2004.
Operational measurement of precipitation in cold climates. In: Meischner P. (ed.), Weather radar. Principles and advanced applications, Springer-Verlag, Berlin,
Germany, pp. 110–114.
Korhonen V.V. 1942. Die Verteilung der Niederschläge,
besonderes der Schneefälle auf die verschiedenen Windrichtungen in Finnland. Ann. Acad. Sci. Fenn. Ser. A1
Math.-Phys. 13: 1–54.
Korhonen V.V. 1944. Ein neues Verfahren bei der Korrektion
der Schneemessungen. Ann. Acad. Sci. Fenn. Ser. A1
Math.-Phys. 24: 1–14.
Lapin M. & Šamaj F. 1991. III: Methods of correcting for
systematic errors in atmospheric precipitation measurements in Czechoslovakia. Hydrol. Process. 5: 243–250.
Legates D.R. & Willmott C.J. 1990. Mean seasonal and spatial variability in gauge-corrected, global precipitation.
Int. J. Climatol. 10: 111–127.
Legates D., Yang D., Quiring S., Freeman K. & Bogart T.
2005. Bias adjustment to arctic precipitation: a comparison of daily versus monthly bias adjustments. 8th
Conference on polar meteorology and oceanography.
American Meteorological Society, Boston MA (USA).
Available at https://ams.confex.com/ams/Annual2005/
techprogram/paper_86285.htm and on CD-ROM.
Metcalfe J.R., Ishida S. & Goodison B.E. 1994. A corrected
precipitation archive for the Northwest Territories.
In: Stewart J.C. (ed.), Mackenzie Basin Impact Study,
23
Interim Report 2, Proc. Sixth Biennial AES/DIAND
Meeting on Northern Climate, Yellowknife, North West
Territories, April 10–14, 2014, Environment Canada,
pp. 110–117.
Michelson D.B. 2004. Systematic correction of precipitation
gauge observations using analyzed meteorological variables. J. Hydrol. 290: 161–177.
Milton J.S. & Arnold J.C. 1987. Probability and statistics in
the engineering and computing sciences. McGraw-Hill,
Inc., Singapore.
Nitu R. & Wong K. 2010. CIMO survey on national summaries of methods and instruments for solid precipitation
measurement at automatic weather stations. Instruments
and observing methods. Report No. 102. WMO/TD No.
1544, World Meteorological Organization, Geneva.
Peel M.C., Finlayson B.L. & McMahon T.A. 2007. Updated
world map of the Köppen-Geiger climate classification.
Hydrol. Earth Syst. Sci. 11: 1633–1644.
Pirinen P., Simola H., Aalto J., Kaukoranta J.-P., Karlsson P.
& Ruuhela R. 2012. Climatological statistics of Finland
1981–2010. Reports 2012:1, Finnish Meteorological
Institute, Helsinki, Finland.
Rasmussen R., Baker B., Kochendorfer J., Meyers T., Landolt S., Fischer A.P., Black J., Theriault J., Kucera P.,
Gochis D., Smith C., Nitu R., Hall M., Cristanelli S. &
Gutmann E. 2011. How well are we measuring snow?
Bulletin of the American Meteorological Society, doi:
10.1175/BAMS-D-11-00052.1.
Saltikoff E., Lopez P., Taskinen A. & Pulkkinen S. 2015.
Comparison of quantitative snowfall estimates from
weather radar, rain gauges and a numerical weather prediction model. Boreal Env. Res. 20: 667–678.
Sarkanen A. 1989. The method for correcting the wind error
of point precipitation measurements. Meteorological
publications no. 13. Finnish Meteorological Institute,
Helsinki, Finland. [In Finnish with English summary].
Schlichting H. 1968. Boundary-layer theory, 6th ed.
McGraw-Hill, New York, USA.
Schwarz K.T., Patzek T.W. & Silin D.B. 2009. Dispersion
by wind of CO2 leaking from underground storage:
Comparison of analytical solution with simulation. Int.
J. Greenhouse Gas Control 3: 422–430.
Sevruk B. 1982. Method of correction for systematic error
in point precipitation measurement for operational use.
WMO No. 589, WMO, Geneva.
Sevruk B. (ed.) 1986a. Correction of precipitation measurements. ETH/IAHS/WMO Workshop on the Correction
of Precipitation Measurements, Zurich 1–3 April 1985.
Geographisches Institut, Eidgenössische Technische
Hochschule, Zürich, Switzerland.
Sevruk B. 1986b. Correction of precipitation measurements.
Summary report. In: Sevruk B. (ed.), Correction of Precipitation Measurements ETH/IAHS/WMO Workshop on
the Correction of Precipitation Measurements, Zurich
1–3 April 1985, Geographisches Institut, Eidgenössische
Technische Hochschule, Zürich, Switzerland, pp. 13–23.
Sevruk B. (ed.) 1989a. Precipitation Measurement. WMO/
IAHS/ETH Workshop on Precipitation Measurement. St.
Moritz, 3–7 December 1989. Swiss Federal Institute of
Technology, Zurich, Switzerland.
24
Sevruk B. 1989b. Reliability of precipitation measurements.
In: Sevruk B. (ed.), Precipitation Measurement. WMO/
IAHS/ETH Workshop on Precipitation Measurement. St.
Moritz, 3–7 December 1989, Swiss Federal Institute of
Technology, Zurich, Switzerland, pp. 13–19.
Sevruk B., Ondras M. & Chvila B. 2009. The WMO precipitation measurement intercomparisons. Atmos. Res. 92:
376–380.
Solantie R. 1986. A quantitative model for point precipitation
correction by use of data from standard meteorological
stations. In: Sevruk B. (ed.), Correction of Precipitation
Measurements ETH/IAHS/WMO Workshop on the Correction of Precipitation Measurements, Zurich 1–3 April
1985, Geographisches Institut, Eidgenössische Technische Hochschule, Zürich, Switzerland, pp. 197–202.
Solantie R. 1987. Precipitation and snow cover. In: Alalammi
P. (ed.), Atlas of Finland. Folio 131, Climate, National
Board of Survey and Geographical Society of Finland,
Helsinki, Finland, pp. 7–8.
Solantie R. & Pirinen P. 2006. Consideration of the orographical effect in the areal analysis of October–March
precipitation. Reports no. 2006:8, Finnish Meteorological Institute, Helsinki, Finland [In Finnish with English
summary and figure captions].
Solantie R., Drebs A., Hellsten E. & Saurio P. 1996. Timing
and duration of snow cover in Finland 1961–1993.
Meteorological publications no. 34, Finnish Meteorological Institute, Helsinki, Finland [In Finnish with English
summary].
Stisen S., Højberg A.L., Troldborg L., Refsgaard J.C., Christensen B.S.B., Olsen M. & Henriksen H.J. 2012. On
the importance of appropriate precipitation gauge catch
correction for hydrological modelling at mid to high latitudes. Hydrol. Earth Syst. Sci. 16: 4157–4176.
Sugiura K., Ohata T. & Yang D. 2006. Catch characteristics
of precipitation gauges in high-latitude regions with high
winds. J. Hydrometeorol. 7: 984–994.
Tammelin B. 1984. On the wind error of the Finnish raingauge. Meteorological publications no. 37, Finnish
Meteorological Institute, Helsinki, Finland. [In Finnish
Taskinen & Söderholm • BOREAL ENV. RES. Vol. 21
with English summary].
Vehviläinen B., Huttunen M. & Huttunen I. 2005. Hydrological forecasting and real time monitoring in Finland: The
watershed simulation and forecasting system (WSFS).
In: Balabanis P., Lumbroso D. & Samuels P. (eds.),
Innovation, Advances and Implementation of Flood
Forecasting Technology, Conference Papers, Tromsø,
Norway, 17–19 October 2005, available at http://www.
actif-ec.net/conference2005/proceedings/PDF%20docs/
Session_03_Pre-operational_flood_forecasting/Vehvilainen_Bertel.pdf.
Wagner A. 2009. Literature study on the correction of precipitation measurements. FutMon C1-Met-29(BY), Bayerische Landesanstalt für Wald und Forstwirtschaft.
Wolff M., Isaksen K., Brækkan R., Alfnes E., Petersen-Øverleir A. & Ruud E. 2013. Measurements of wind-induced
loss of solid precipitation: description of a Norwegian
field study. Hydrol. Res. 44: 35–41.
Yang D. 1999. An improved precipitation climatology for the
Arctic Ocean. Geophys. Res. Lett. 26: 1625–1628.
Yang D. & Ohata T. 2001. A bias-corrected Siberian regional
precipitation climatology. J. Hydrometeorol. 2: 122–139.
Yang D., Goodison B.E. & Metcalfe J.R. 1998. Accuracy
of NWS 8´´ standard nonrecording precipitation gauge:
Results and application of WMO intercomparison. J.
Atmos. Ocean. Tech. 15: 54–68.
Yang D., Ishida S., Goodison B.E. & Gunther T. 1999.
Bias correction of daily precipitation measurements for
Greenland. J. Geophys. Res. 104(D6): 6171–6181.
Yang D., Goodison B.E., Metcalfe J.R., Golubev V.S.,
Elomaa E., Günther Th., Bates R., Pangburn T., Hanson
C.L., Emerson D., Copaciu V. & Milkovic J. 1995.
Accuracy of Tretjakov precipitation gauge: result of
WMO intercomparison. Hydrol. Process. 9: 877–895.
Ye B., Yang D., Ding Y., Han T. & Koike T. 2004. A biascorrected precipitation climatology for China. J. Hydrometeorol. 5: 1147–1160.
Zhang Y., Ohata T., Yang D. & Davaa G. 2004. Bias correction of daily precipitation measurements for Mongolia.
Hydrol. Process. 18: 2991–3005.